基于动态贝叶斯网的航班延误传递分析  

Analysis of flight delay transmit based on dynamic Bayesian network

作　　者：丁建立[1] 赵键涛 曹卫东[1] 胡海生[2] 黄威[2] 

机构地区：[1]中国民航大学计算机科学与技术学院,天津300300 [2]中国民航信息网络股份有限公司,北京100010

出　　处：《计算机工程与设计》2015年第12期3311-3316,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(60879015);中国民航科技重大专项基金项目(MHRD201241);中央高校基本科研业务费基金项目(3122014P004)

摘　　要：为对航班延误传递的特性进行研究,采用动态贝叶斯网络对延误传递过程进行分析。在动态贝叶斯网络结构确定的条件下,不同的参数学习方法效果不同,以似然函数作为度量准则,比较两种不同的参数学习方法在相同数据集下的学习结果,在较好的学习结果上,对航班延误传递的特性进行分析。分析结果表明,当上游航班发生延误时,若飞机过站时间充分,大多数情况下延误可被吸收,为航班延误传递的关键环节过站时间的设置提供了很好的参考;航班延误传递是一个非稳态过程,当延误在传递的过程中时,延误时间的不确定性是逐渐增大的,预测准确的可能性是逐渐减小的。To study the property of flight delay transmission, dynamic Bayesian network was used to analyze the flight delay propagation. Different parameter learning methods have different outcomes when the structure of dynamic Bayesian network is established. Two methods of learning the parameter applied on the same dataset were compared using the likelihood as criterion. The features of flight delay transmit were analyzed based on the good result. The result indicates that in most cases, when up- stream flight is delayed, the delay can be absorbed when the turnaround time is sufficient, which provides reference for airline companies setting up turnaround time which is the key to flight delay transmit. Besides, the flight delay transmit is not a steady process, the uncertainty over delay time is increasing while the delay is transmitting, so the possibility of accurate prediction is decreasing.

关 键 词：航班延误传递 动态贝叶斯网 参数学习 EM算法 梯度优化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]